A housing adviser opens a regulation that has been amended six times. The question is ordinary and urgent: which wording applied when the tenant received the notice? A search result may show the latest text, a PDF may preserve an earlier edition, and a consolidated version may be easier to read. If those objects are not clearly identified and connected, a useful digital service can produce a precise answer to the wrong version of the law.
Machine-readable legislation is meant to reduce that risk. Its most valuable function is not to turn law into software that decides cases. It is to give legal texts stable identities, explicit structure and traceable relationships so that people and systems can find the right provision, date and source. The boundary matters: structured data can improve navigation and verification, but legal authority and interpretation must remain attached to an accessible human-readable text and the institution responsible for publishing it.
A document needs more than a filename
A conventional digital document is often a container of pages. A person can infer that a bold line is an article heading, that a smaller block is a paragraph and that a phrase refers to another act. A computer may see only coordinates, characters or visual formatting. It can search the words, but it does not reliably know what role they play.
Structured legislation makes those roles explicit. An article is represented as an article, a paragraph as a paragraph, and a reference as a reference rather than as a run of styled text. Metadata can identify the issuing body, language, type of act, date, jurisdiction and legal status. Relationships can connect an original act to an amendment, corrigendum, consolidated presentation or later version.
This creates a useful division of labour. Machines can retrieve every provision with a particular identifier, check whether a link resolves, compare versions or warn that a service is using an outdated expression. People can read the provision in context, examine its legal source and decide what it means in a particular situation.
ELI gives law an address and a history
The European Legislation Identifier, or ELI, is a framework maintained through the Publications Office of the European Union. Its first layer gives legislation web identifiers assembled from components chosen by the official publisher. A URI can distinguish the kind of act, year, number, version and language. The address is meaningful to a reader but can also be followed by software.
ELI then adds a common data model for metadata. It distinguishes the abstract legal work from particular expressions and formats, helping a system separate a regulation from its English expression and from a specific HTML or PDF file. The framework supports publication of this metadata in RDFa or JSON-LD embedded in legislative web pages. Its newer synchronisation specification lets data users retrieve complete datasets and updates rather than repeatedly guessing what has changed.
The result is not a universal European numbering system imposed on every legal order. Official publishers retain their own URI patterns and responsibilities. ELI supplies compatible concepts and relationships so that a legal research tool can traverse sources across institutions and countries without pretending that they are governed by one database.
Akoma Ntoso gives the text an internal skeleton
Where ELI chiefly identifies and describes legal resources, Akoma Ntoso provides an XML vocabulary for the contents of parliamentary, legislative and judicial documents. OASIS approved version 1.0 as a standard in August 2018. Its elements can represent familiar legal units, metadata and references in a form that different institutions and tools can exchange.
The distinction between appearance and meaning is central. A printed heading can look correct even if it has been created with spaces and bold type. A structurally encoded heading belongs to a defined provision, carries an identifier and can be addressed independently. That makes it possible to generate accessible views, create dependable links to a provision and compare two versions at the level of legal units instead of treating both documents as undifferentiated strings.
Structure does not eliminate interpretation. A reference may be ambiguous. An amendment may affect several provisions in ways that require legal analysis. A definition may depend on context rather than syntax. XML can record the decision made by a drafter or editor; it cannot prove that the decision is legally correct.
LEOS brings structure into drafting
Structure is most reliable when it is created during drafting rather than reconstructed after publication. The European Commission’s LEOS project is an open-source web tool for collaborative legislative drafting, version control and co-editing. It uses Akoma Ntoso and the AKN4EU profile developed for interoperability between EU institutions and Member States.
A drafter works through an editing interface while the system preserves a structured representation underneath. Templates and constrained editing can prevent some formatting and numbering errors; annotations and version comparison can make review more visible. The gain is not simply faster publication. When the drafting record, structure and published formats share a controlled source, fewer fragile conversions stand between a legal change and the reader who needs to understand it.
That does not mean every drafting constraint is substantively neutral. A template can privilege the cases its designers anticipated. A rigid schema can make an unusual but lawful form difficult to express. Good legislative tooling therefore needs escape routes, expert governance and a visible record of transformations rather than silently forcing the text to fit the software.
Metadata errors can be legally consequential
Machine readability moves some risk from typography into metadata. A wrong commencement date, language code or link between an amendment and its target can be propagated through search engines, public-service portals and commercial legal databases. Valid XML only proves that a document follows technical rules; it does not prove the factual accuracy of the encoded relationships.
Publishers need automated validation and human editorial checks. ELI provides a validator for conformance to its ontology, but legal quality also requires reconciliation with the publication workflow, audit trails, correction notices and monitoring of downstream updates. A service should show the source, version date and status close to the answer, not hide them in technical metadata that only developers can inspect.
The authority boundary is especially important. Under Council Regulation (EU) No 216/2013, the electronic edition of the EU Official Journal is authentic and produces legal effects, subject to the exceptional publication arrangements in that Regulation. By contrast, EUR-Lex states that its consolidated texts are documentation tools without legal effect. Both may be expertly structured and extremely useful. Their legal status is nevertheless different because authority comes from the publication regime, not from the data format.
People still need a readable path to the source
The strongest objection to structured law is that it can create a second, specialist layer controlled by technical teams while citizens receive a simplified rendering they cannot verify. The answer is not to retreat to scanned pages. It is to make every machine-generated service point back to a stable, accessible human view, disclose the version being used and preserve the authentic source.
Accessibility must be tested at the rendered level. Semantic structure can support navigation by headings, meaningful links and alternative formats, but only if the publication interface exposes it correctly. A perfectly modelled XML file behind an inaccessible viewer does not improve effective access to law.
This is where the human-machine partnership is most useful. Machines can maintain links, flag broken provenance, assemble version histories and make legislation searchable across languages and institutions. Human publishers remain responsible for the legal classification, corrections and accessible presentation; lawyers, officials and citizens retain the work of interpretation and contestation.
If you value practical examinations of where public technology should assist rather than decide, you can subscribe to Alkemata for future articles in this series.
The next decision is therefore institutional rather than merely technical: will a government treat structured legislation as a transparent public infrastructure tied to authoritative texts, validation and correction, or as an invisible data feed whose errors become everyone else’s problem? Machines can read the law more usefully only when people can still see which law they are reading, where it came from and why it should be trusted.